AI investment is booming. It's time to test investors' ability to think independently.

With 80 exhibitor companies and over 150 robot products on display, in line with the audience's perception, the hottest sector at this year's WAIC, whether in terms of tech startups or investment topics, is undoubtedly embodied intelligence.

"As an investor, I'm a bit nervous," said Zhou Zhifeng, the managing partner of Qiming Venture Partners, at the venture capital forum of WAIC on July 28.

Just in the past month, there have been frequent financing news from several embodied intelligence companies. Large enterprises and institutions are vying to make investments, and hot money is pouring in. Everyone wants to get a seat at the table, driving up the valuations of leading companies. According to IT Juzi data, as of now, there have been 99 investment and financing events in the domestic humanoid robot field this year, far exceeding the 67 events for the whole of last year. However, this sector is still full of high uncertainty.

After joining Qiming Venture Partners in 2014, Zhou Zhifeng's first project was to invest in Megvii Technology, which was one of the "Four AI Unicorns" at that time, together with Kuang Ziping, the founding managing partner of Qiming Venture Partners. In the new wave of AI fever, Qiming has also invested in startups such as Zhipu AI and Jieyue Xingchen, which still adhere to the path "towards AGI". Looking back, 2014 marked the technological singularity of deep learning, and 2020 was the technological singularity of the "GPT - 3 moment". Since then, the industry has had a common direction to strive for.

Regarding the investment in embodied intelligence today, Zhou Zhifeng said that there are still many discussions within Qiming Venture Partners about whether the technology of embodied intelligence has reached a singularity and started to converge or is still in a state of diverse development. After all, the latter means great uncertainty for investors.

The bad news is that the technology of embodied intelligence has not started to converge. Almost every issue, from software and hardware to implementation scenarios, remains open. However, the good news is that practitioners have reached a certain consensus at the macro - level, such as the importance of data and the end - to - end model deployment form. Meanwhile, the technological development of embodied intelligence in the past year has almost exceeded the accumulation of all previous years, and the progress is accelerating.

The AGI industry has reached a special development stage. On the one hand, the technology is still growing rapidly without an obvious ceiling in sight. At the same time, as the performance and cost of the technology become more viable, large - scale applications have begun to be implemented.

Since last year, institutions have felt that AI investment has been getting "hotter". In the first half of 2025, AI startups attracted 53% of global venture capital funds. Although there have been arguments like "the path of pre - training is almost at an end, and the Scaling Law may not work anymore", investment still continues to flow into basic model companies.

53% of global venture capital funds flow into AI startups / Photo by Jiemian News reporter

The "hot" AI investment also means more noise. How to make independent judgments and plan strategically amidst the noise is a test for investors.

From the perspective of entrepreneurs, AI startups consume a huge amount of resources and are among the most fiercely competitive industries globally. Starting a business in such an industry is also becoming more difficult.

Yin Qi, the CEO of Megvii Technology, had a conversation with Kuang Ziping about his recent insights on AI entrepreneurship in his new capacity as the chairman of Qianli Technology.

"When I started my business in 2011, it was the boom of college student entrepreneurship. There was a saying that 'entrepreneurship is like jumping off a cliff and then assembling an airplane'. However, AI is the most competitive industry. At least, there should be a basic design from technology to product to commercialization in entrepreneurship. If the basic elements of the business model are missing or incorrect, all efforts may be in the wrong direction. Therefore, I believe that a business model that cannot form a closed - loop cannot be sustained," said Yin Qi.

Regarding the trends and opportunities for entrepreneurship in the AI 2.0 era, Yin Qi believes that it is very difficult to start a lightweight pure - APP business. In the era of mobile Internet, many entrepreneurial legends emerged by breaking through in niche areas. However, the large - model industry is a well - known and high - stake field, which means there are few opportunities for surprise attacks.

"The rapid flow of information and talent drives the quick spread of technology application concepts. There is no time window for startups to break through," said Yin Qi. However, he believes that there are still huge opportunities in the hardware field. There are definite opportunities in the automotive, smart home, and other major areas, as well as more high - quality opportunities in vertical hardware segments.

During the conversation, Kuang Ziping also revealed his thoughts as an investor. At the forefront of this wave of AI, when investing in technology platform companies, the goal of the institution is not just to invest in a single hardware product like the "coolest AI printer", but to invest in the most core AI entry - end hardware platform.

Qiming Venture Partners believes that the shift of AI interaction paradigms will accelerate in the next two years. As users rely less on mobile phone screens and the importance of natural interaction methods such as voice increases, it will drive the birth of AI - native super - applications.

53% of global venture capital funds flow into AI startups / Photo by Jiemian News reporter

Back at WAIC, the Future Tech section features 50 startups taking turns to present their projects, showcasing the latest AI innovations. There are also investment and financing matchmaking events to promote innovation.

At this exhibition area, the CEO of a startup told Jiemian News that there has been a continuous stream of investors during this year's WAIC. There are even overseas capital groups coming to learn about the situation. This is in line with the globalization of AI. In the past, Internet companies started locally and then expanded overseas, but AI products may be "born global" from the very beginning.

Qiming Venture Partners believes that in the next one to two years, a 2 - million - Token context window will become the standard for top - tier AI models. More refined and intelligent context engineering centered around larger context windows will become one of the core driving forces for the development of AI models and applications. In terms of multimodality, a general video model is expected to emerge in one to two years. This model can handle generation, reasoning, and task understanding in the video modality, promoting innovation in video content generation and interaction.

Regarding the currently popular AI Agents, Qiming Venture Partners predicts that in the next one to two years, the form of Agents will shift from "tool assistance" to "task undertaking". The first batch of genuine "AI employees" will enter enterprises and be widely involved in core processes such as customer service, sales, operations, and R & D. They will no longer just serve as assistants but will have the capabilities of collaborative work, proactive feedback, and taking on OKRs, driving the transformation from cost - based tools to value creation.

The AI BPO (Business Process Outsourcing) model will also achieve commercial breakthroughs in the next one to two years, shifting from "delivering tools" to "delivering results". Through the "pay - by - result" approach, it will rapidly expand in industries with standardized processes such as finance, customer service, marketing, and e - commerce.

Embodied intelligence robots will be the first to achieve large - scale deployment in scenarios such as picking, handling, and assembly. They will accumulate a large amount of first - person view data and tactile operation data, constructing a closed - loop flywheel of "model - body - scenario data". This flywheel will drive the iteration of model capabilities and ultimately promote the large - scale implementation of general robots.

This article is from "Jiemian News", author: Cheng Lu, editor: Wen Shuqi. Republished by 36Kr with permission.